Developer Chat

smortierIRC Welcome to today's Science Chat everybody!! 15:00
bkoep Hi everyone! 15:01
@phi16 hello smortier, TvdL 15:01
smortierIRC Today, we are going to mostly focus on the questions that were posted by players in the SciChat announcement, but if you want clarification or have a quick, easy-to-answer question, feel free to ask and we'll try to address it over this next hour. 15:01
smortierIRC We'll start by introducing the admin team very quickly. I'll begin with myself: 15:01
smortierIRC My name is Saira and I am Foldit's Community Liaison. My job is to be the players' voice to the rest of the admin team and vice versa. I'll let the rest of the team hop in and introduce themselves. 15:02
bkoep I'm Brian, a member of the Baker Lab at the University of Washington 15:02
@jflat06 I'm Jeff (jflat06) - I work on the server and client software as a member of the Computer Science Department at the University of Washington. 15:03
@jflat06 (i guess my username is already in the message line, doh!) 15:03
rmoretti Hello, my name is Rocco, and I'm in the Meiler Lab at Vanderbilt University, and I'm focusing on developing the small molecule design portions of Foldit. 15:03
horowsah Hi all, I'm Scott, and I'm an Assistant Professor at the University of Denver, working mostly on educational aspects of Foldit at the moment. 15:04
ianh Hi - I'm Ian, a scientist working in the Baker lab at the University of Washington. I work on enzyme design full time :) 15:05
smortierIRC Great! Now let's give into those player questions on the Science Chat announcement post. 15:05
smortierIRC Question 1: Is an entry point needed for ligand? You mentioned a paper with crocodile jaws allowing the entry of a drug or ligand. I wonder if an entry point is obligatory, or is it possible that a protein forms and surrounds completely a ligand? Then with no entry or escape point. 15:06
bkoep This is a good question! The paper in question is the Nature paper by Polizzi et al. 15:06
bkoep 15:06
bkoep It turns out that entry and exit paths are not as important as you might think. 15:07
bkoep In Foldit, we tend to focus on a protein’s single most stable structure (the native state). This is the state in which the protein spends most of its time, and tells us the most about the protein’s stability and function. 15:07
bkoep But we have to remember that, in aqueous solution at room temperature, proteins are very dynamic molecules. They are constantly unfolding and refolding, to some degree, which can expose or occlude a binding pocket. 15:07
bkoep Still, accessibility can certainly affect ligand binding kinetics. If a protein is so stable that it very, very rarely exposes a hidden binding pocket, then ligand-binding will be very slow (but not necessarily weak!). 15:08
bkoep In the study by Polizzi et al., they did not design their protein to open and close, per se. Rather, they designed a binding pocket deep in the protein core; as a result, the protein core is under-packed when the ligand is absent, making it less stable and more likely to flex. (Remarkably, another region of the protein core remains well-packed and rigid, both 15:08
bkoep ... with and without a ligand present 15:08
smortierIRC Great. Does anyone have any follow up questions on this? 15:09
@TimovdL In the latest design puzzles more interacton with the ligand is not rewarded extra. I feel those interaction should have more weight if you want solutions with more interaction. 15:10
@MikeCassidytoo Yes I found that puzzling 15:10
bkoep @Timo, that's a good point, and it's something we've experimented in the past 15:10
@MikeCassidytoo Some of the solutions that out scored me had no extra bonds 15:11
bkoep We wouldn't want to overdo it (because changing the weighting for a single residue will affect the balance of other residues) 15:11
@TimovdL Just a little bit more might give you more interactione 15:11
bkoep But we may try upweighting ligand interactions in future puzzles 15:12
smortierIRC Alright. Moving on to question 2. 15:12
smortierIRC Question 2: Ligand/drug destruction by protein. In the aflatoxin example, I understand that a protein could destroy or deactivate a ligand by bonding to it. Is it the case for drugs? (protein destroying the drug)? 15:12
rmoretti Certainly! In fact, metabolism of drugs by enzymes is a major consideration during drug design. There are large classes of liver enzymes (Cytochrome P450s) which are responsible for “detoxifying” drugs. 15:13
rmoretti If compounds are too reactive to P450s, they don’t work well as drugs. Alternatively, things like grapefruit can inhibit P450 enzymes, reducing anticipated metabolism and resulting in “too much” drug being present. 15:13
smortierIRC Thanks, Rocco! Question 3: Can the scientists take us through the expected changes to the aflatoxin? 15:14
bkoep This is a great question, but I don’t think I can answer it satisfactorily in chat 15:15
bkoep I’m going to ask the Siegel Lab if they will put together a blog post with some details about the strategy for degrading aflatoxin. 15:15
@MikeCassidytoo TY 15:16
bkoep I don't know if @ianh will have anything to say at this time? 15:16
ianh I'm working on making a figure that will hopefully make the changes to aflatoxin clear. I'll post a link to the picture here as soon as it's ready. Just a minute or two... 15:17
smortierIRC Great. I'll move on to question 4 in the meantime. 15:17
smortierIRC Question 4: In regards to playerbase demographics, when was the last count taken for this sort of thing? From what I know the playerbase right now tends to be older and I don't see too many of my peers folding yet. Some updated numbers could be useful for scoping out the state of the game. 15:18
smortierIRC This question in more in my wheelhouse, so I'll take it. 15:18
smortierIRC Once a year, we ask players to fill out a demographics survey. We have not yet sent one out for 2017 but it is in the works. Age of players tends to be fairly diverse and evenly split. 15:18
smortierIRC Question 5: Can we please know who determines what proteins are going to the lab, if not all the protein options? Is it possible for a citizen scientist to gain credit for their designs? 15:20
bkoep We collaborate with a number of research groups and scientists. The results of any puzzle are typically analyzed by those most familiar with the project. 15:21
bkoep I can claim responsibility for running the analysis of recent Foldit design puzzles, and for determining which Foldit player designs will be tested in the Baker Lab. 15:21
bkoep Concerning "credit": If any of your Foldit solutions are reported in a Foldit paper, or are used in research elsewhere, then those solutions will be attributed to you. 15:22
bkoep If you are looking for academic credits from the University of Washington, you will have to enroll at the university and take accredited classes like any other student. 15:22
bkoep Playing Foldit may be a great way to familiarize yourself with topics in biochemistry, but merely participating in Foldit puzzles hardly meets the standards for undergraduate education. 15:22
bkoep Foldit puzzles are no substitute for a rigorous study of biochemistry. 15:22
@TimovdL Let everybody know who made those and if the designs fail why 15:23
ianh ok - back to the question about how aflatoxin changes one the enzyme reacts with it. I love this question! Enzymes are my specialty 15:23
ianh here's how I think the chemistry is going to work: 15:24
ianh that link should take you to a picture I just tweeted. 15:24
ianh Now, in plain english: 15:24
ianh In step 1, the enzyme binds to aflatoxin. There is water nearby – because proteins live in water –  but at this point the water and the aflatoxin are not interacting. 15:24
ianh In step 2, the enzyme “splits” the water (H2O -> H+ and OH-). The OH- part then “attacks” the lactone group. This destabilized aflatoxin but does not break the lactone group completely. 15:24
ianh In step 3, the H+ part of water finishes the reaction. This is where the lactone ring is finally broken. After this step, the enzyme is ready to catalyze another reaction. 15:24
smortierIRC Follow up question for Ian? 15:25
@jeff101 are free radicals involved? 15:25
@MikeCassidytoo A above my 60 yr old high school chem but very cool 15:25
ianh nope - this is acid/base catalysis. Some enzymes use free radicals, but this one doesn't need em 15:26
bkoep @Timo, if you're asking for more feedback about Foldit player designs that make it to the lab, I apologize that I haven't put together another blog post about those designs 15:26
@Susume2 what role does the calcium play? 15:26
ianh Calcium is what is needed to split the water into the H+ and OH- parts. Other side chains are likely helping that happen as well 15:27
ianh That process of splitting a water molecule in half is super common in enzyme chemistry. We call it "activating" a water molecule. Enzyme often need a metal ion to help make it possible 15:28
@jeff101 is it important for the enzyme to protect the reaction from water and other side reactions with solvent 15:28
ianh Yes and no – it's complicated. The reaction actually requires water in order to happen at all, but too much water can interfere. There is going to have to be balance 15:29
@TimovdL Are there specific sidechains that would help doing the splitting, if so any near the ligand should also be extra rewarede 15:29
ianh It's probably more important to have the enzyme bind aflatoxin near the metal. 15:30
ianh Without that binding, nothing can happen. Solve that and you're well on your way to breaking the lactone ring 15:31
rmoretti The specific sidechains which help with the splitting are already in the starting structure and are frozen in the puzzle. 15:31
ianh ^ yep. These are excellent questions! 15:32
smortierIRC Okay. We'll move on to question 6: I understand that the Sketchbook goal is to encourage us to work more manually, and trying more candidate designs. It sounds like you would be able to identify a good candidate design on day 4. Is there a way to help the players (incl. evolvers) to select the best "science" candidate after several days of semi-manual design? 15:32
rmoretti Everything that's needed for the core of the reaction is there ... the only thing missing is the enzyme's ability to bind aflatoxin in particular. 15:32
@Susume2 is bonding to the lactone O itself good or bad? 15:32
ianh probably good 15:32
ianh that's the best I can say :) enzyme are tricky things 15:33
bkoep For the Sketchbook question: The following answer will be somewhat oblique, but I think addresses the main concerns behind this question 15:33
bkoep Some of us suspect that late game strategies do not improve the scientific merit of Foldit models—especially in prediction puzzles. 15:33
bkoep The scoring function used by Foldit and Rosetta is only a rough approximation of protein physics, and the score function can be abused. 15:34
bkoep With perseverance (and some clever scripting) it’s possible for a player to wring out incremental score gains from a perfectly good model without making meaningful changes to the model. 15:34
bkoep We fear that this practice has become common among competitive Foldit players, even though—from a scientific perspective—it is a waste of resources that would be better spent exploring different, alternative models. 15:34
@TimovdL I know 15:34
bkoep Of course, all of that is hypothesis. The Sketchbook puzzles are part of an experiment to test that, restricting late game strategy by limiting player move counts. 15:35
@MikeCassidytoo That is very true 15:36
smortierIRC Question (Suggestion) 7: Learning from other players & Chat time & date stamp. For improving my own folding skills I would like to see - from each 'expired' puzzle - the best ranking solution loadable into my solution e.g. as a reference/guide - full colored. chat should also print date & time (after an ongoing chat paused for a certain time). My laptop might run the whole day, but I'm not sitting the whole day in front of t 15:36
smortierIRC he screen. 15:36
@jflat06 I really like the idea of letting the community's experience help to educate newer players. 15:37
@MikeCassidytoo BUT it is also part of the *game* - us competing for two or three pts to best each other 15:37
@jflat06 That said, there are some potential issues with this: 15:37
@TimovdL But, as I am a scriper I will finc a way to get around the move restrictions 15:38
@jflat06 Often times we will run puzzles in series, and showing other players' results from previous puzzles might give away some of their hard work. 15:38
@jflat06 We want the purpose to be an educational advantage, not a method for using other player's work. 15:39
@jflat06 So in order for this to work, we would need some system for selectively allowing loading in top-scored solutions. 15:39
@jflat06 A reasonable compromise might just be any puzzle older than a couple of months, as these rarely are part of continued series. 15:40
@jflat06 For the time being, though, this feature will have to take a back seat to other priorities. 15:40
smortierIRC Question 8: When bonding to a ligand, is it better that there be no bindable but unbound sidechains? For example, if a GLN fits and binds just right with only one arm, is that a bad bond compared to another AA that has only one bindable sidechain? 15:41
@jeff101 regarding the sketchbook puzzles, mine the data we send you for normal puzzles so you don't ignore lower-scoring solutions 15:41
bkoep @jeff101, rest assured, we pay careful attention to all Shared with Scientist solutions! 15:42
@jflat06 @Mike, @Timo - the hope is that the best way to "compete" switches from squeezing a few points out of your existing solution, to trying a new track/trajectory 15:42
bkoep @question_8, Yes! 15:43
@MikeCassidytoo i would love blue print for sketchbook puzzles 15:43
@jeff101 Share with Scientists is only 5 solutions ... we send multiple solutions per hour that we play 15:43
bkoep It is very important that every polar atom either (1) makes a hydrogen bond or (2) is exposed on the protein surface, so that it can form hydrogen bonds with water. 15:43
bkoep This is a rather important feature in protein structure that is perhaps underemphasized in Foldit. Good question! 15:43
@Susume2 in the frozen parts of the aflatoxin puzzle are some sidechains with unbonded red bondable atoms - shoudl we be trying to bond those? (I don't actually think we can reach them...) 15:44
@Susume2 the frozen sidechains that are already bonding the aflatoxin 15:44
bkoep I wouldn't worry about the frozen sidechains. Especially near the calcium ion, those oxygens may be satisfied even if they don't make explicit hydrogen bonds to other protein residues. 15:45
@Susume2 131N, 188N 15:45
@Susume2 ok 15:45
smortierIRC Question 9: Could you give us a quick update on foldit designs? How many have passed the Rosetta test, how many have passed the wet lab tests, how many have been crystallized (even if not yet solved)? Are the successful ones all the same shape? (I heard a little about this after the aflatoxin launch, but I know others want to know too). 15:47
bkoep I don’t have a thorough answer prepared for this question, but the highlights are: 15:48
bkoep A few months ago we ordered about 75 designs by Foldit players. A large number of them appear to be structured and stable in solution, but we’d really like crystal structures. 15:48
bkoep We have diffraction data from two or three protein crystals that we are trying to solve. Hopefully we’ll have new structures soon! 15:48
smortierIRC Question 10: Is there anything to report on a possible current effort or progress being made in rebalancing or refining the accuracy of foldit scoring internally as opposed to externally by using filters? 15:49
@jflat06 If you mean refinements to the score function itself - yes! Rosetta developers are continually working on this and making some great updates. 15:50
@jflat06 In Foldit, we avoid taking these updates too often, because it tends to be disruptive when the score function is shifting around underneath you. 15:50
@MikeCassidytoo cool 15:50
@jflat06 So we don't always have the latest-and-greatest. 15:50
@jflat06 That said, we will be updating our score function soon to get up-to-date with these improvements. 15:50
smortierIRC Question 11: Can we please have a feature to STOP the rebuild when You identify the rebuild you want(while its searching in the purply colour), before it moves on to a new attempt to find new energy 15:51
rmoretti Sounds like a reasonable idea. If there isn’t one already, please make a Feedback entry for it. 15:51
rmoretti We do pay attention to feedback entries, and are deeply appreciative about the bug reports and comments you have. That said, we’re a small team and we can’t always get to addressing issues as rapidly as we would like. (Comments and votes on Feedback entries help us prioritize.) 15:52
@TimovdL It might be better to update more as amall changes wouldn't be disruptive as one big change as long performance is not to much degrading 15:52
@jflat06 @Timo, that somewhat depends on the state of the Rosetta source 15:52
@Susume2 I think the purply rebuilds are ones that foldit can't close the cut - when it manages to close the cut it goes in the undo graph - jflat might know 15:53
@jflat06 We care most about getting updates when Rosetta has a new default score function, which isn't something we can smoothly transition into when it happens. 15:53
@jflat06 @Susume, I believe that is correct 15:53
@jflat06 But if I'm understanding the question - would an interface like Remix be more approriate? 15:54
@LociOilingIRC I'd vote for that 15:54
@TimovdL I would love to see that 15:54
@MikeCassidytoo Yes 15:54
@Susume2 I liek the remix interface much better than the rebuild 15:54
@MikeCassidytoo We could step through them 15:54
@jflat06 I'm not sure how much work we're going to do on Rebuild at this point, given that we've had some data suggest it results in poorer fragment qualities. 15:55
@jflat06 It can be very useful for things like tail-ends, though. 15:55
@jflat06 and when remix doesn't work 15:55
@Susume2 but it's so good for point wringing ;-) 15:55
@jflat06 haha 15:55
@TimovdL This is one I have already on my list 15:55
@MikeCassidytoo ;-) 15:55
smortierIRC Question 12: Re: Helix bias. I've read that the Foldit optimizer has some bias in favour of helices. Is this bias random? If not, why can't you correct it with some filter penalty? 15:56
bkoep There does seem to be some bias in Rosetta and Foldit for alpha-helices—especially in protein design. But that “bias” not entirely unrealistic. 15:56
bkoep Alpha-helices are exceptionally stable structures, due to their regularly spaced hydrogen bonds (like in sheets) and short range side-chain interactions (unlike sheets). 15:57
@TimovdL This should really be something for the Rosetta score function 15:57
bkoep In design puzzles, we do in fact have to correct for this, using the Secondary Structure Filter. Otherwise, helical bundles dominate the rankings. 15:57
bkoep Even when we're not always interested in helical bundles 15:57
crpainter Are cys-cys bonds undervalued in the Rosetta score function? 15:58
bkoep @crpainter, that's a difficult question to answer. 15:59
crpainter In many puzzles you add a 250 point bonus for such bonds 16:00
@TimovdL Are othe helix types overpenalized by the Rosetta score functions is also something I am thinking of 16:00
Enzyme2 and without the filter bonus, Making disulfide bonds is detrimental to the score 16:01
bkoep Because Cys-Cys disulfide bonds form a covalent bond, they behave differently from other physical forces 16:01
bkoep In particular, they have a big effect on the entropy of the unfolded state, which we often are unable to account for in Rosetta modeling 16:02
bkoep So, when we know disulfides should be present, we make a point of adding the filter to encourage players to form them 16:03
smortierIRC Question 13: For membrane proteins, orange should be outside. Do you correct the scoring function for these "lipidic" (or partially lipidic) puzzles? 16:04
rmoretti Right now support for membrane proteins in Foldit is limited. The big issue is correctly determining where the membrane “should” be. 16:04
rmoretti The underlying Rosetta code can do this, but there’s issues with the Rosetta/Foldit interface which currently make it challenging to do in Foldit, which means we’re not currently making the corrections we should be for those puzzles. 16:05
rmoretti There are updates in the pipeline, though, which should address this. (No ETA, however.) 16:05
smortierIRC Question 14: Now we can design drugs. Once you have a drug design, how can you produce this exact drug in the labo? 16:06
rmoretti That’s the question we’re asking ourselves. :) 16:06
rmoretti You’re absolutely correct that producing a small molecule in the lab is more complex than producing a certain amino acid sequence. We’re currently looking into various computational methods to predict how one might make small molecules that Foldit players design. 16:06
rmoretti There’s already the “synthesizability” filter, which gives a rough estimate of how hard something is to make in the lab. 16:06
rmoretti We’re also looking into other techniques to both help guide players to make small molecules which are easier to make in the lab, and also to help the synthetic chemists figure out how to make them. 16:07
@TimovdL That filter should be give a bonus for easy to make 16:08
rmoretti @Timo - it should do so already 16:09
smortierIRC Question 15: Aflatoxin 1445. The reverse (grayed/frozen) side of the aflatoxin grabber seems to have the structure of an ideal funnel trap to let the aflatoxin enter or even be 'sucked' in. If that's true and the funnel side-chains don't clog the passage, what we need to do is to block the rear entrance and 'grab' the aflatoxin in the 'pocket' like a baseball in a glove to keep if from getting through or backing out. What do you think? 16:09
bkoep Blocking the passage is probably not so important. We talked a little bit about this at the very beginning of the chat, but we’re not so concerned with ligand entry/exit in a ligand-binding protein. 16:09
bkoep If you design a pocket that binds the ligand very strongly, then the ligand will prefer to be in the pocket. 16:10
@TimovdL And you allready explained that the protein is probably flexible without the ligand so blocking is not a real issue for entering 16:11
ianh There is also a slight concern that if you build the protein so that it grabs aflatoxin on all sides and never lets go, you won't have made a very good enzyme. It has to be able to bind to and eventually let go of the molecule 16:12
@TimovdL So the glove should not be too perfect 16:13
ianh exactly 16:13
rmoretti That's a good problem to have, though -- chances are designs will under bind rather than over-bind. 16:13
rmoretti (And it's easier to mess up binding than it is to make it stronger.) 16:14
smortierIRC Looks like our time is up! Thank you everyone for your questions. 16:14
smortierIRC As a reminder, chats are logged and we post the log afterwards. 16:14
smortierIRC Thanks, everyone! 16:14
bkoep Thanks all for the great questions! 16:14
@jflat06 Thanks! 16:15
crpainter Thank you! 16:15
@LociOilingIRC thanks to ianh for joining us today! 16:15
ianh my pleasure - it was fun chatting with you all 16:15
@LociOilingIRC and great "epoxy" blog post... 16:15
ianh Thanks Loci 16:15
ianh see you! 16:16
@jeff101 thanks 16:17

Generated by 2.13.1 by Marius Gedminas
- find it at!

Get Started: Download
  Windows    OSX    Linux  
(10.7 or later)

Are you new to Foldit? Click here.

Are you a student? Click here.

Are you an educator? Click here.
Only search
Recommend Foldit
User login
Top New Users

Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, RosettaCommons